Nova Southeastern University Nova Southeastern University

NSUWorks NSUWorks

All HCAS Student Theses, Dissertations, and Capstones HCAS Student Theses and Dissertations

8-5-2020

Algal Coverage Detection and Classification using ENVI: Algal Coverage Detection and Classification using ENVI:

Correlation with Dissolved Oxygen Levels in Elkhorn Slough, CA Correlation with Dissolved Oxygen Levels in Elkhorn Slough, CA

Jason Dawson Nova Southeastern University

Follow this and additional works at: https://nsuworks.nova.edu/hcas_etd_all

Part of the Aquaculture and Fisheries Commons, Environmental Indicators and Impact Assessment

Commons, Environmental Monitoring Commons, Oceanography Commons, and the Other Environmental

Sciences Commons

Share Feedback About This Item

This Capstone has supplementary content. View the full record on NSUWorks here:

https://nsuworks.nova.edu/hcas_etd_all/2

NSUWorks Citation NSUWorks Citation Jason Dawson. 2020. Algal Coverage Detection and Classification using ENVI: Correlation with Dissolved Oxygen Levels in Elkhorn Slough, CA. Capstone. Nova Southeastern University. Retrieved from NSUWorks, . (2) https://nsuworks.nova.edu/hcas_etd_all/2.

This Capstone is brought to you by the HCAS Student Theses and Dissertations at NSUWorks. It has been accepted for inclusion in All HCAS Student Theses, Dissertations, and Capstones by an authorized administrator of NSUWorks. For more information, please contact nsuworks@nova.edu.

Capstone of Jason Dawson

Submitted in Partial Fulfillment of the Requirements for the Degree of

Master of Science Marine Science

Nova Southeastern University Halmos College of Arts and Sciences

August 2020

Approved: Capstone Committee

Major Professor: Dr. Bernhard Riegl

Committee Member: Dr. Mara Orescanin

This capstone is available at NSUWorks: https://nsuworks.nova.edu/hcas_etd_all/2

NOVA SOUTHEASTERN UNIVERSITY

HALMOS COLLEGE OF ARTS AND SCIENCES

ALGAL COVERAGE DETECTION AND CLASSIFICATION USING

ENVI: CORRELATION WITH DISSOLVED OXYGEN LEVELS IN

ELKHORN SLOUGH, CA.

By

Jason M Dawson

Submitted to the Faculty of

Halmos College of Arts and Sciences

in partial fulfillment of the requirements for

the degree of Master of Science with a specialty in:

Marine Environmental Sciences

Nova Southeastern University

August 5, 2020

1

TABLE OF CONTENTS

I. ABSTRACT ............................................................................................................5

II. INTRODUCTION: ................................................................................................6

A. BACKGROUND ........................................................................................6

1. Estuary Definition:.........................................................................6

2. Human Impact on Estuarine Systems: ........................................7

3. Eutrophication Impact: .................................................................9

B. STUDY SITE:.............................................................................................9

1. Hypothesis: ...................................................................................12

III. FIELD AND ANALYSIS METHODS ...............................................................13

A. IMAGERY ACQUISITON .....................................................................14

B. IMAGE ANALYSIS PROCESSING .....................................................16

1. ENVI Calculations of Algae Coverage .......................................17

2. ESNERR Calculations of Algae Coverage .................................18

IV. RESULTS/DISCUSSIONS .................................................................................19

A. IMAGERY ANALYSIS ..........................................................................19

1. January 16 – Field Survey ...........................................................21

2. January 19 – Field Survey ...........................................................22

3. July 18 – Field Survey..................................................................23

4. October 18 – Field Survey ...........................................................24

B. WQ SEASONAL ANALYSIS .................................................................25

C. IMAGERY AND WQ DISCUSSION.....................................................28

V. CONCLUSIONS ..................................................................................................31

LIST OF REFERENCES ................................................................................................34

2

LIST OF FIGURES

Figure 1. Restoration of the Nisqually Delta (diked) ..................................................8

Figure 2. Elkhorn Slough, Watsonville, CA .............................................................10

Figure 3. DO measurements collected from North Marsh (NM), ES. ......................11

Figure 4. WQ sampling locations within ES. ............................................................13

Figure 5. DJI Phantom 4 Pro Quadcopter. ................................................................15

Figure 6. Jan 16 Original RGB vs ENVI classification ............................................21

Figure 7. Jan 19 Original RGB vs. ENVI classification ...........................................22

Figure 8. Jul 18 Original RGB vs ENVI classification .............................................23

Figure 9. Oct 18 Original RGB vs ENVI classification ............................................24

Figure 10. 2016 Wet Season Trend .............................................................................26

Figure 11. 2018 Dry Season Trend .............................................................................27

Figure 12. 2019 Wet Season Trend .............................................................................28

Figure 13. Band Manipulation Features with ENVI ...................................................32

3

LIST OF TABLES

Table 1. Flight Survey and Instrument Inventory ....................................................15

Table 2. Image Classification (ENVI/DroneDeploy) comparison with Avg DO. ...19

Table 3. Error Matrix for Jan 16 ..............................................................................21

Table 4. Error Matrix for Jan 19 ..............................................................................22

Table 5. Error Matrix for Jul 18 ...............................................................................23

Table 6. Error Matrix for Oct 18 ..............................................................................24

Table 7. ENVI Algal Classification Accuracies ......................................................29

4

LIST OF ACRONYMS AND ABBREVIATIONS

ABW Above the water

AC Acres

CDMO Central Data Management Office

Chl-a Chlorophyll-a

DJI Da Jing Innovations

DO Dissolved Oxygen

ES Elkhorn Slough

ESNERR Elkhorn Slough National Estuarine Research Reserve

IAW In accordance with

MP Megapixels

RGB Red Green Blue color spectrum

SAM Spectral Angle Mapper

SAL Salinity

SAV Submerged Aquatic Vegetation

SQFT Square feet

TEMP Temperature

UAV Unmanned Aerial Vehicle (drone)

UW Underwater

WQ Water Quality

5

I. ABSTRACT

Estuaries are exposed to varying stressors, whether they be physical, chemical, or

environmental. The most notable of stressors is eutrophication of coastal and inland

ecosystems. This is a result of increased supply of nutrients fueling production within the

system. One outcome of this increased nutrient load to the system is that of algal blooms.

These blooms can impact the aesthetic appearance and degrade the quality of health of the

system. Many of these coastal zones and waterways are critical habitats for many biological

(some endangered) species and serve as recreational areas for human populations. Elkhorn

Slough, California is one of these critical habits. Over its history, land use and environmental

changes have degraded the quality of the ecosystem. Elkhorn Slough National Estuarine

Research Reserve (ESNERR) has been tasked with oversight and monitoring responsibilities

to maintain the system at a suitable level for the native species to thrive. This study, in

conjunction with ESNERR support, will use aerial imagery of designated restoration areas to

investigate the ability to use spectral analysis techniques to identify, classify, and calculate the

percent coverage of algae masses. The aim is to use the inherent spectral analysis toolboxes

in Harris Geospatial’s ENVI to ingest 3-band RGB imagery and differentiate and accurately

classify algal coverage. The goal is to compare ENVI’s performance and accuracy, using

ground-truthed base-image against traditional, time-intensive hand analytics. There is an

extensive imagery library that has not be analyzed. This study will assess the potential ability

to automate the process and increase classification capabilities.

Keywords: Estuaries, Water Quality, Environmental Monitoring, Spectral Analysis, Elkhorn

Slough, algae monitoring, dissolved oxygen, hypoxia, eutrophication

6

II. INTRODUCTION

A. BACKGROUND

Estuaries are remarkable systems common to global coastlines and are considered some

of the most biologically productive environments in the world (Wise 2017). They provide

ecologically and economically valuable wildlife refuge, biological nursery, and dynamic

nutrient transformation zones (Chuwen et al 2009; Hoeksema et al 2018; Paerl et al 1998).

Three-quarters of the world’s population resides within coastal river basins; forty-four percent

of that population lives within 150 km of the coast, and this percentage is even higher within

the United States (Handler et al 2006; Le et al 2011; Paerl et al 1998). Increased awareness

of, and attraction to these zones has created a dependence upon these systems for sustainment

and recreation.

1. Estuary Definition

The term estuary is defined as a system characterized by and emphasis on riverine input

and at least a periodic connection with the ocean; where some regions are described by their

geometry of permanently open to the ocean, where others are closed from the ocean by a sand

bar across their mouths, either intermittently, seasonally, or for a protracted period of time

(Hoeksema et al 2018). These coastal zones are areas where terrestrial rivers meet ocean

influences, impacting its ecological state or nature (Palmer et al 2011 and Plew et al 2015).

Classification of estuaries varies according to different aspects, while the diversity of estuary

types is well known from literature, the major contributors defining type are the relationships

of tidal range, sediment status, and relative wave-tide-fluvial processes (Cooper 2001). Global

coastlines vary in all the above-mentioned factors. Therefore, classification of an estuary may

be difficult, each has a unique combination of tidal, climate, geomorphological, and river input

(Palmer et al 2011). Traditionally, the classification has been attributed to the coastal geometry

and the stratification owing to seawater intrusion (Hansen and Rattray1966). Independent of

classification, estuaries have been and remain vital to commercial, and recreational industries;

for the purposes of tourism, fisheries, and water supply (Le et al 2011).

7

2. Human Impact on Estuarine Systems

Estuaries have great capacity to support and maintain biodiversity (terrestrial & marine)

and provide protection to coastal infrastructure by buffering energetic wave activity from the

ocean. These low-profile systems provide a barrier to extreme flooding events associated with

storms and hurricanes (Leonardi et al 2018). They are also the focal points for urban,

industrial, and agricultural development and usage as recreational activities (Hoeksema et al

2018). Exchanges and interactions from increasing populations within coastal zones tend to

have consequences, at times; significant. Large population densities and impacts of human

activities on coastal ecosystems can result in various alterations such as deterioration of water

quality (WQ) and significant changes in the hydrological and biogeochemical cycles and

biodiversity (Guimaraes et al 2017), to name a few.

The physical layout of estuaries can be altered as well. The protective nature and

energy-buffering capabilities provided creates a nutrient rich system within the sediments and

encompassing water, adding to the economic worth and attraction (Kennish 2019). Flow

patterns have been changed in many systems and estuaries have been diked to use land space

and sediments for agricultural purposes (Figure 1) (Orescanin et al 2019). These expansions

in estuarine land use alter critical hydrological and biogeochemical dynamics, impacting health

and biological suitability further up or down stream, and historically dried out areas subside

too low for marsh to survive (Clark and O’Conner 2019). A secondary affect from coastal

wetland diking is the significant degradation of WQ throughout the system due to a lack of

flushing (water residence time) behind tidal and flow control structures (Orescanin et al 2019).

8

Figure 1. Restoration of the Nisqually Delta (diked)

Population overcrowding and additional environmental stressing has the potential to

negatively impact the system. The past few decades, research and studies have resulted in an

increased focus in gaining a better understanding of the physical and biochemical balances that

maintain “suitable” and “good” health of our global estuaries. It is of world-wide concern that

estuaries are now considered the most degraded of all temperate marine ecosystems (Chuwen

et al 2009). While increased regulatory efforts in the United States and elsewhere have yielded

many improvements in environmental quality compared to a few decades ago, environmental

degradation persists on a global scale and broad environmental goals have yet to be achieved

(McCormick and Cairns 1994). Government agencies and the general public have become

increasingly concerned about maintaining the quality of aquatic resources (Maznah and Omar

2010). The ability to protect biological resources depends on the ability to identify and predict

the effects of human actions on biological systems; thus, the data provided by indicator

organisms can be used to estimate the degree of environmental impact and its potential danger

for others (Karr and Chu 1999).

9

3. Eutrophication Impact

Eutrophication is one of the main WQ issues worldwide, affecting both freshwater and

marine ecosystems (Guimaraes et al 2017). Eutrophication is defined as the supply of excess

nutrients and supplements that result in excessive growth of primary producers. Plankton and

microalgae can contribute up to 95% of the primary production distribution in response to

readily available nutrients (Kotsedi et al 2012). All facets of a eutrophic system can and will

be negatively impacted within the estuarine system, if not properly monitored and kept in

check. Eutrophication of surrounding coastal waters is a problem of epidemic proportion and

can have disastrous short- and long-term consequences for WQ and resource utilization (Millie

et al 2004). Twomey and Thompson (2001) found that unmonitored eutrophic systems would

progress to a level of biological and hydrologic degradation due to the emergence of harmful

algal blooms (HAB’s), which can result in the total loss of habitat biota, and commercial value.

Human activities compound the task of monitoring and assessing health status of

estuaries. Resultant land alterations can increase nutrient loads, displaying nuisance algal

blooms, degraded WQ, and habitat loss (Twomey and Thompson 2001). As Agencies and the

general public have become more aware of the potential dangers associated with hydrologic

flow alterations, significant time and effort has been put into understanding the primary

production and algal concentrations (Geyer et al 2018; Guimaraes et al 2017; Handler et al

2006; Hoeksema et al 2018; Hubertz and Cahoon 1999; Kotsedi et al 2012; Maznah and Omar

2010; McCormick and Cairns 1994; Millie et al 2004; Sharp 2010; Paerl et al 1998; and Paerl

et al 2007). Much of this work has focused on phytoplankton populations rapid response to

physical, chemical, and biological changes in eutrophic estuarine systems (Hsieh et al 2007).

B. STUDY SITE

Elkhorn Slough (ES) is a 7-mile-long tidal estuary located in central California that is

permanently open to Monterey Bay (Figure 2). It’s a biologically rich environment that

provides habitat diversity to resident and migratory birds, plants, marine mammals, and fish,

and has been identified as a Globally Important Bird Area by the American Birding

Conservancy, and recently received a RAMSAR classification for significant estuarine habitats

(ESNERR 2019).

10

A: Aerial Image of the extent of ES., B: Location of ES in relation to Monterey Bay, CA., and C:

Location of ES in relation to the California Coast.

Figure 2. Elkhorn Slough, Watsonville, CA

This ecological and economically valuable ecosystem, like many others across the

globe, has been and is being threatened by local human impacts, such as agriculture and energy

development. Local farming activities are responsible for introducing excess nutrients and

other harmful pollutants that exacerbate biological and geochemical changes within the

system. Eutrophication in estuaries surrounding these agriculturally and productive sites

commonly suffer from the phenomena of hypoxic and anoxic conditions, due to the excessive

algal growth and production from readily available nutrients (Bricker et al 2007). Alterations

of the biogeochemical cycling can have long lasting impacts. ES has experienced episodic

periods of hypoxic (DO < 2 milligrams per liter (mg/L)), and anoxic (DO = 0 mg/L) conditions

within the various arms and branches of the estuary (Figure 3). Aside from the stresses of

hypoxia and anoxia, DO levels greater than 5 mg/L are considered oxic and “suitable for

supporting biological diversity. DO levels between 2 – 5 mg/L are a transitional zone, leading

to a hypoxic environment.

11

Figure 3. DO measurements collected from North Marsh (NM), ES.

ES supports hundreds of species of fish, invertebrates, and birds, and is considered one

of the most extensive salt marshes in the state outside of San Francisco Bay (Jeppesen et al

2016). With this perspective, ES has been the focus of WQ monitoring and restoration efforts.

Over the years of monitoring, ES has developed a report card based on WQ conditions.

Monitoring sites within the Lower ES region have better WQ relative to the southern estuary

sites and those located in the northern-most ES reaches, attributing the differences to existing

water control structures (Mercado et al 2014). The range of daily average DO values was from

0.2 to 14 mg/L, with values dipping into the hypoxic zone in the summer months (Jun – Aug)

(Schmit 2010). Biological and physical factors impact the variability of DO levels resulting

from excess algal and plankton growth, and tidal exchange (residence times). In addition to

those previously mentioned, restricted circulation in estuaries is now being considered a major

factor. The majority of sites with poor WQ and increased hypoxia corresponded to those

located behind these water control structures that restrict flow and increase the residence time;

sites with better quality were close to the mouth and areas along the lower channel, which had

unrestricted flow, full tidal exchange, and short residence time (Mercado et al 2014).

12

1. Hypothesis

Estuarine systems are the focal areas of WQ restoration efforts and monitoring

programs. These complex environments have dynamic interactions that can alter, degrade and

ultimately, result in habitat and species loss. Investigations into health maintenance and

indicators of degradation are abundant with focuses on using algae as the primary variable

(McCormick and Cairns 1994). Algae blooms can alter a system chemically and physically,

and most notably deplete DO levels in the surface waters (Egerton et al 2014). Techniques use

Chlorophyll-a (Chl-a) as a proxy for phytoplankton biomass, a primary response to

eutrophication that leads to WQ degradation (Stanley 1993). A review of these studies, past

and present, DO was not considered as a primary WQ parameter. Salinity (SAL), Temperature,

Chl-a, and nutrients have primary emphasis placed on them as directly related to WQ

conditions and health (Kotsedi et al 2012). without the need for hand-annotated estimates of

algal coverage.

The objective of this research is to determine the reliability of using the ENVI to

identify and classify algal (Ulva [sp]) coverage from 3-band RGB imagery collected remotely,

within the NM site of ES. Furthermore, algae coverages estimated will be investigated to

determine possible fluctuations in the DO seasonal trend. The hypotheses to be tested here are

that 1) Harris Geospatial’s ENVI spectral analysis program can be used to differentiate and

accurately classify algal coverage (against verified base-imagery), automating algal coverage

calculation within estuarine environments; and 2) Algae coverage can be used as an indicator

of DO levels.

Image classification stats will be compared against the current hand-analysis technique,

as a faster and accurate alternative process available for use.

13

III. FIELD AND ANALYSIS METHODS

This study was performed as a partnership with the Elkhorn Slough National Estuarine

Research Reserve (ESNERR), to assess algal growth relationships with WQ at the four (4)

sites within ES that are considered areas of concern for ecological impacts due to

eutrophication (Figure 4). Identification of any correlations between the percent cover of algae

and the area of an estuarine system is one that requires information from two main sources, 1

– WQ measurements (DO, SAL, temperature), and 2 – photographic images algae growth

within the estuary over time. ESNERR collected aerial imagery over the NM site over the past

four years and provided the images from their survey database. WQ data that was downloaded

from the Central Data Management Office (CDMO) website. The images were initially

classified by ESNERR using DroneDeploy and hand annotation. Here, images are classified

using the spectral analysis program, ENVI to separate and calculate area coverage. That

coverage was then compared to the corresponding time series of WQ data to identify trends or

correlations between algae growth and changes in the DO and duration of hypoxia.

Figure 4. WQ sampling locations within ES.

14

Past studies have used various methods for algae coverage calculation, ranging from

in-situ sampling to using Unmanned Aerial Vehicles (UAVs) as remote sensing surveys, to

looking for the spectral signatures of chlorophyll-a (Chl-a) (Koparan et al 2018, Hansen et al

2017, Le et al 2013, Palmer et al 2013, Ali 2011, Egerton et al 2014, and Reif 2011). Matthew

and Bernard (2013), like many of the previous studies found that Chl-a was a strong indicator

and could be used as a proxy for algal and phytoplankton biomass. Their research found that

Chl-a correlated to spectral reflectance in the 420 – 480 nm range. Other pigment

characteristics found specifically to algae were found to have signatures in the 620 – 700 nm

range, lending to the applicability of using remote sensing techniques and applications to

quickly and accurately assess large areas of coastal waters (Matthew and Bernard 2013).

For this study, algal cover is estimated through spectral classification using ENVI with

the imagery collected by UAV’s during field studies. A recent study showed that ENVI offers

multiple capabilities for classification, using various algorithms and regions of interest (ROI’s)

dictated by the user, and is accurate in identifying specific features or interest (Mielke 2019).

ENVI, a spectral analysis program was selected to classify provided imagery for is ease and

speed of automatic classification process

A. IMAGERY ACQUISITON

The ESNERR team started collecting and building a photographic library of ES in late

2015. These image surveys are conducted monthly or whenever feasible to access the site,

using a UAV with a mounted camera. The Da-Jiang Innovations (DJI) Phantom 4 Pro

quadcopter outfitted with a 20-megapixel (MP) camera (Figure 5) has been used to collect the

3-band (RGB color spectrum) images. Flight plans were preset and executed in an east-west

direction with a nadir-looking orientation. Individual images are then stitched together using

photogrammetry techniques, creating a full coverage ortho-mosaic image of the sample

location. All was accomplished using the mapping software application DroneDeploy.

Mission planning used a 75% front overlap with a 65% side overlap. Flights were flown at 60

meters (m) with a single pass and no perimeter oblique views. Historical weather and

environmental conditions were collected, corresponding to each survey (Table 1) in order to

account for variability in lighting.

15

Table 1. Flight Survey and Instrument Inventory

Image Jan 16 Feb 16 May 16 Apr 18 Jul 18 Sep 18 Oct 18 Jan 19 Mar 19

Date taken: 12 10 02 09 25 17 25 23 04

Imaging

Sensor:

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

1/2.3"

CMOS

Flight time: 1202 1351 1044 1325 1356 1322 1408 1258 1410

Wind speed

(m/s): 2.2 4.0 3.1 5.4 1.3 4.9 1.3 2.2 3.6

Cloud

cover: Clear Clear Cloudy

Partly

cloudy Cloudy Clear Clear Clear

Mostly

cloudy

Pixel size

(m): 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04 0.04

Meteorological data recorded at the time of flights. All times are local times (PST/PDT). GMT =

+8/+7, respectively. Data was collected from the Elkhorn WX station (KCAWATSO38) (Weather

Underground 2020). From here on, all images will be reference with the title listed here (month, year).

Figure 5. DJI Phantom 4 Pro Quadcopter.

Orthomosaic maps were processed and geo-rectified with DroneDeploy software using

fixed ground control points which remained constant for each flight, then exported as geotiffs

for use in ENVI. Nine (9) 3-band RGB geotiffs were provided for classification analysis in

DroneDeploy (ESNERR) and ENVI (this study). These images are from the NM location

16

(Figure 5). Images analyzed here spanned from 2016 – 2019 and were taken across a variety

of months.

B. IMAGE ANALYSIS PROCESSING

Image processing used ENVI for spectral signature identification and classification, similar to

studies conducted using satellite imagery for coral reef habitat mapping and tracking of

bleaching events in that remote imagery is processed (Riegl and Purkis 2005; Rowlands et al

2008). ENVI provides different methods of automatic classification through its use supervised

and unsupervised internal algorithms (Mielke 2019).

ENVI’s classification accuracy was dependent upon consistency throughout the

process with all nine images. Image normalization was performed for assessment of the

spectral classification capability and helped with comparison of classified outputs. Every ROI

file consisted of the same five class types for feature identification. The included classes used

a standard color scheme during classification (purple = dirt/mud, cyan = water, yellow =

vegetation, bright green = algae (abw), and dark green = algae (uw)).

ENVI’s classification process requires that the program have a set of “class types”

composing ROI files for analysis and classification of remotely collected imagery. Ideally

these ROI’s are verified to what the environmental types are at a set time. For this study, 5

classes; algae (abw), algae (uw), dirt/mud, vegetation (other than algae), and water were

identified as the prominent features across all images. Using the ROI function, each base

image was used to ROI build files containing the spectral signatures of the 5 classes. Each of

the class types is identified by drawing a polygon around features the correspond to the class.

ROI files containing the desired class types was saved as an overlay to be used in classification.

This overlay serves as ground-truth for post-analysis error matrix and total coverage

calculations. Class features selected in the ROI were verified during the field surveys by visual

inspection from prominent features that remained constant across seasonal changes (increase

in water level, etc.). Algae (abw and uw) coverage was the only variable feature but was

evident and recognizable in all images. This in-situ verification of the base images will be the

basis, assessing classification performance of ENVI. ROI files were required for each image

because there was no anchor point, linking all images to a common projection.

17

Next step was to analyze each image with one of the various classification methods

provided. The supervised Spectral Angle Mapper (SAM) was the choice algorithm during the

classification process because of the nature of the algorithm to identify spectral signatures

within the image. As previously mentioned, ENVI contains various classification algorithms.

ENVI support documentation was reviewed to confirm the applicability to SAM to goals of

this study (Harris1 2020). Supervision refers to the fact that training data or the generated ROI

files is required for classification. Following the prompts within ENVI, each base image was

used as the input file and compared against is corresponding ROI file. Completing the

classification process, the output file will be the representation of the base image, colored in

the scheme dictated by the ROI files and their class representation color.

Post- classification work included the creation of the error matrix comparing ENVI’s

producer accuracy to its user accuracy performance assessment. Producer accuracy measures

ENVI’s ability of correctly identify the algae classes, where user accuracy represents the

probability measure of ENVI to place class type values (or pixels) into the correct class

grouping. (Harris 2020). These matrices are generated internal to ENVI, in the confusion

matrix toolbox. Tables 3 through 6 provide the accuracy assessment of ENVI’s prediction

capability for this study.

After the completion of the classification process and error matrix generation,

quantification of the algal extent in the images was needed. Both ENVI and DroneDeploy had

different procedures of calculating the overall percent coverage. The methods of both analysis

tools for calculating total percentage are:

1. ENVI Calculations of Algae Coverage

Internal to ENVI’s classification protocol, a text file (.txt) is generated with all the

spectral information per classified image. The file contains class type name, area coverage (in

pixel count and percent) of the image. Total algal coverage for each image was derived from

the addition of both algae (abw and uw) class counts. Total percent coverage is annotated in

Table 2.

18

2. ESNERR Calculations of Algae Coverage

Coverage calculations are performed in the web-based program DroneDeploy. This

program requires using hand analysis of the imagery, marking and designating areas of interest,

like using Google Earth (DroneDeploy 2020). Classification is conducted drawing polygons

around visual areas of identifiable algal patches. Each polygon covers image pixels and

internally converts those pixels to a corresponding area in acres (AC). All generated polygons

areas were summed, then divided by the total marsh area (roughly 100.98 AC), yielding total

coverage.

19

IV. RESULTS/DISCUSSIONS

A. IMAGERY ANALYSIS

The technique of using optical remote imagery for identification and habitat mapping

has been documented as a suitable method within shallow water systems (Rigel and Purkis,

2005). Based on that premise, images collected from a UAV were collected and classified

using ENVI and DroneDeploy. The classification performance of ENVI was evaluated and

results were compared against the “hand-analysis” technique in DroneDeploy. Algal

identification and coverage calculations were the primary focus of both methods and the basis

of comparison. Table 2 shows resulting coverage comparisons between respective methods

and measured DO levels at the time the nine (9) images were collected.

Table 2. Image Classification (ENVI/DroneDeploy) comparison with Avg DO.

IMAGE DATE ENVI Spectral

Calculation

Drone Deploy

Calculation Difference

DO Level

(Avg)

JAN 16 16.3 14.5 +1.8 NA

FEB 16 18.5 14.7 +3.8 7.5

MAY 16 23.9 29.0 -5.1 3.5

APR 18 13.6 8.0 +5.6 4.5

JUL 18 14.0 2.8 +11.2 3.8

SEP 18 4.8 1.0 +3.8 2.8

OCT 18 13.0 4.9 +8.1 5.9

JAN 19 15.6 15.3 +0.3 8.1

MAR 19 17.1 11.5 +5.6 9.2

ENVI, DroneDeploy, and difference values are coverage percentages for algae. DO values are

mg/L (average for the collection date of the image). Blue rows - wet season and orange rows - dry

season.

Total algal coverage (percent) varied between the two method capabilities. Overall,

relative skill comparison between ENVI and DroneDeploy are very close with an RSME of

4.3%, indicating a good agreement in capability to identify and differentiate desired features.

Comparisons shown in Table 2 represent estimates of the desired algae from a given images

per each method, respectively. Four (4) images (two with smallest difference and two with the

largest difference) are used to show the accuracy of ENVI and its automated analysis for

brevity. These images will be referenced by the collection month and year (i.e., Jan 16, Jan

20

19, Jul 18, and Oct 18). Figures 6 through 9 are comparisons of ENVI classified outputs against

the original 3-band RGB images.

Tables 3 through 6 represent the accuracy assessment of ENVI’s prediction

capabilities, compared to ground-truth data. ENVI was very precise with is classification

protocol, effectively identifying greater than 90% of algae (both classes) coverage within the

Jan16 images (Figure 6). The Jan 19 (Figure 7) similarly, posted similar results within the

producer category which is ENVI’s ability of correctly identify the algae classes (Harris 2020).

There was a significant disparity in the user accuracy for Jan 19 (Figure 7). User accuracy

represents the probability measure of ENVI to place class type values (or pixels) into the

correct class grouping. (Harris 2020). ENVI performed at 51%, placing the features with

similar pixel properties int to the algae (uw) class. Despite the confusion with submerged algae

in the Jan 19 image (Figure 7), ENVI displayed excellent accuracy. Total coverages for Jan

16 (Figure 6) and Jan 19 (Figure 7) comparing ENVI’s automation process against

DroneDeploy hand-analysis were very close, at 1.8% and 0.3%, respectively (Table 2).

For the Jul 18 (Figure 8) and Oct 18 (Figure 9) images, ENVI was efficient in its

producer classification capability, classifying algal types, with accuracies ranging from 96%

to 98% (Table 5 and Table 6). Significant errors occurred in the discrimination of algae (uw)

pixels, with an accuracy at 8.2%. Oct 18 (Figure 9) did not see the same disparity that was

exhibited in Jul 18 (Figure 8). Algae (uw) attribution was at 87%, with these variations

possibly contributing to the total coverage differences of +11.2 and +8.1, respectively (Table

2).

21

1. January 16 – Field Survey

Cyan coloring indicates water. Yellow represents vegetations such as marsh grass (growing

in the middle of the marsh). Light and dark shaded green indicates algae. Trees and

shrubbery were classified in this grouping as well. Purple indicates exposed dirt and

submerged mud regions. ENVI and DroneDepoly total algae was 16.3% and 14.5%,

respectively (Table 2).

Figure 6. Jan 16 Original RGB vs ENVI classification

Table 3. Error Matrix for Jan 16

CLASSES - GROUND TRUTH

(PSEUDO)

Dirt and

Mud Water Vegetation

Algae

(UW)

Algae

(ABW)

CL

AS

SE

SS

-

PR

ED

ICT

ED

Dirt and Mud 85.3 0.0 5.2 0.0 0.0 87.3

Water 8.0 100.0 3.9 0.4 9.1 96.2

Vegetation 6.7 0.0 88.1 0.0 0.0 96.9

Algae (UW) 0.0 0.0 0.0 99.6 0.0 99.9

Algae (ABW) 0.0 0.0 2.8 0.0 90.9 98.2

TOTALS 100.0 100.0 100.0 100.0 100.0

US

ER

AC

C.

PROD. ACC 85.3 100 88.2 99.6 90.9

Note - Values were generated using the ENVI Confusion Matrix toolbox. All values are

percentages

22

2. January 19 – Field Survey

Cyan coloring indicates water. Yellow represents vegetations such as marsh grass (growing

in the middle of the marsh). Light and dark shaded green indicates algae. Trees and

shrubbery were classified in this grouping as well. Purple indicates exposed dirt and

submerged mud regions. ENVI and DroneDepoly total algae was 15.6% and 15.3%,

respectively (Table 2).

Figure 7. Jan 19 Original RGB vs. ENVI classification

Table 4. Error Matrix for Jan 19

CLASSES - GROUND TRUTH (PSEUDO)

Dirt and

Mud Water Vegetation

Algae

(UW)

Algae

(ABW)

CL

AS

SE

SS

-

PR

ED

ICT

ED

Dirt and Mud 91.0 0.1 21.6 0.0 0.7 0.6

Water 0.0 97.4 2.9 0.7 0.4 92.2

Vegetation 9.0 1.6 70.1 0.1 0.2 98.3

Algae (UW) 0.0 0.5 5.4 93.8 2.1 51.4

Algae (ABW) 0.0 0.4 0.0 5.4 96.6 99.6

TOTALS 100.0 100.0 100.0 100.0 100.0

US

ER

AC

C.

PROD. ACC 91.0 97.4 70.1 93.8 96.6

Note - Values were generated using the ENVI Confusion Matrix toolbox. All values are percentages

23

3. July 18 – Field Survey

Cyan coloring indicates water. Yellow represents vegetations such as marsh grass (growing

in the middle of the marsh). Light and dark shaded green indicates algae. Trees and

shrubbery were classified in this grouping as well. Purple indicates exposed dirt and

submerged mud regions. ENVI and DroneDepoly total algae was 13.9% and 2.8%,

respectively (Table 2).

Figure 8. Jul 18 Original RGB vs ENVI classification

Table 5. Error Matrix for Jul 18

CLASSES - GROUND TRUTH (PSEUDO)

Dirt and

Mud Water Vegetation

Algae

(UW)

Algae

(ABW)

CL

AS

SE

SS

-

PR

ED

ICT

ED

Dirt and Mud 94.8 46.9 0.2 0.6 1.0 56.0

Water 4.8 0.1 17.3 1.0 2.5 0.4

Vegetation 0.4 33.9 80.6 0.5 0.0 30.4

Algae (UW) 0.0 19.1 1.9 97.9 0.1 8.2

Algae (ABW) 0.0 0.0 0.0 0.0 96.4 99.9

TOTALS 100.0 100.0 100.0 100.0 100.0

US

ER

AC

C.

PROD. ACC 94.8 0.1 80.6 97.9 96.4

Note - Values were generated using the ENVI Confusion Matrix toolbox. All values are

percentages

24

4. October 18 – Field Survey

Cyan coloring indicates water. Yellow represents vegetations such as marsh grass (growing

in the middle of the marsh). Light and dark shaded green indicates algae. Trees and

shrubbery were classified in this grouping as well. Purple indicates exposed dirt and

submerged mud regions. ENVI and DroneDepoly total algae was 13% and 4.9%,

respectively (Table 2).

Figure 9. Oct 18 Original RGB vs ENVI classification

Table 6. Error Matrix for Oct 18

CLASSES - GROUND TRUTH (PSEUDO)

Dirt and

Mud Water Vegetation

Algae

(UW)

Algae

(ABW)

CL

AS

SE

SS

-

PR

ED

ICT

ED

Dirt and Mud 74.6 0.0 18.6 0.0 0.0 72.9

Water 0.0 98.3 2.9 3.0 2.1 96.3

Vegetation 25.4 1.6 76.9 0.0 0.4 80.3

Algae (UW) 0.0 0.1 1.6 96.5 0.3 87.1

Algae (ABW) 0.0 0.0 0.0 0.5 97.2 99.9

TOTALS 100.0 100.0 100.0 100.0 100.0

US

ER

AC

C.

PROD. ACC 74.6 98.3 76.9 96.5 97.2

Note - Values were generated using the ENVI Confusion Matrix toolbox. All values are

percentages

25

B. WQ SEASONAL ANALYSIS

WQ analysis was conducted focusing on the variations within the DO levels, on data

collected from the Central Data Management Office (CDMO), ranging from 2016 - 2020. The

Central California Coast experiences two major seasons – wet, occurring from October to April

and dry, from May to September. The data indicated seasonal trends present in DO

measurements during we seasons where levels reach and maintain normal oxic conditions (DO

> 5 mg/L), with low occurrences of hypoxia. Conversely, during dry seasons, the DO greatly

varies, with more frequent hypoxic events and experiencing period of anoxia (DO < 0.5 mg/L).

Figures 10 to 12 compare the wet and dry seasons for years 2016, 2018 and 2019,

corresponding to images (Figures 6 -9 and Tables 3 - 6).

Trend lines (longer term averages) for both DO measurements and water level were

analyzed for identification and comparison of seasonal trends. Trends were calculated by

applying a seven (7) day moving filter (or moving average) over the data record, attempting to

remove tidal influences and diurnal fluctuations. DO was divided up to identify the varying

conditions. Green markers indicate oxic conditions; blue markers are the transition (2 -5 mg/L)

state; yellow markers indicate hypoxic conditions, and red markers indicate the anoxic

conditions.

26

Time series of water level (cyan) and DO (colored) versus time for the wet season, 2019. Black line

is the average water level, pin line is the average DO concentration. For DO, green markers indicate

oxic conditions; blue markers are the transition (2 -5 mg/L) state; yellow markers indicate hypoxic

conditions, and red markers indicate the anoxic conditions. Red dash lines indicated corresponding

classification imagery.

Figure 10. 2016 Wet Season Trend

27

Time series of water level (cyan) and DO (colored) versus time for the dry season, 2018. Black line is

the average water level, pin line is the average DO concentration. For DO, green markers indicate oxic

conditions; blue markers are the transition (2 -5 mg/L) state; yellow markers indicate hypoxic

conditions, and red markers indicate the anoxic conditions. Red dash lines indicated corresponding

classification imagery.

Figure 11. 2018 Dry Season Trend

28

Time series of water level (cyan) and DO (colored) versus time for the wet season, 2019. Black line

is the average water level, pin line is the average DO concentration. For DO, green markers indicate

oxic conditions; blue markers are the transition (2 -5 mg/L) state; yellow markers indicate hypoxic

conditions, and red markers indicate the anoxic conditions. Red dash lines indicated corresponding

classification imagery.

Figure 12. 2019 Wet Season Trend

Average DO levels ranged between 5 – 10 mg/L, experiencing fewer hypoxic and

anoxic periods throughout the season (Figs 10 and 12). Drier seasons experienced decreased

average DO values, ranging from 2 – 8 mg/L, with significantly more hypoxic and anoxic

occurrences (Figure11). DO variations show a strong agreement with drier months and lower

water levels factoring into poorer WQ conditions, while the wet season and increased water

levels indicated better WQ, respectively. The average DO values ranged from 2.8 – 9.2 mg/L,

across the time range of the four images selected for classification accuracies (Figures 6 – 9

and Table 2).

C. IMAGERY AND WQ DISCUSSION

ENVI proved to be a suitable tool for quick and accurate algae classification from

remotely sensed imagery. Classified Images and error matrices were generated in ENVI for

each image provided. Table 6 shows the average classification accuracy (user and producer)

29

for algae (abw and uw) classes across all images. Overall, ENVI’s precision identifying

specific class types and capability of placing correct features in correct classes is good. Larger

variations in accuracy was seen in the programs ability to handle placement of algae,

specifically submerged (uw) information into the correct class groupings, with an average

accuracy of 58%. Algae on the surface (abw) was handled very well, posting user and producer

accuracies ranging from 94% and 96%, respectively.

Table 7. ENVI Algal Classification Accuracies

USER ACCURACY PRODUCER ACCURACY

Algae (abw) Algae (uw) Algae (abw) Algae (uw)

IMA

GE

12 Jan 16 98.2 99.9 90.9 99.6

10 Feb 16 99.9 63.5 97.4 99.0

02 May 16 100.0 98.7 98.9 98.8

09 Apr 18 71.0 0.3 99.1 91.8

25 Jul 18 99.9 8.2 96.4 97.9

17 Sep 18 91.6 27.6 98.5 77.8

25 Oct 18 99.9 87.1 97.2 96.5

23 Jan 19 99.6 51.4 96.6 93.8

04 Mar 19 89.3 87.3 93.0 79.7

Average 94.4 58.2 96.4 92.8

All values are percentage extracted from the Error Matrices generated at the conclusion of the

classification process in ENVI.

Variations seen in ENVI’s ability to place class type information into the correct class

grouping can be associated to the process routines that look at placing similar image pixels into

similar classifications. Hestir et al (2008) conducted a similar study looking at submerged

aquatic vegetation (SAV) and found that pixel composition of the target species presented a

common problem due to spectral signature blending with surrounding vegetation and like-

features.

Variations in the algae (uw) percentages across images (Table 6) appear to be impacted

by pixel confusion, particularly where the visible algae resembled that of the surrounding

water. Use of spectral analysis identifying vegetation submerged below the water surface tend

to have similar characteristics of the surrounding water (Everitt et al 1999). Similarly, high

amounts of organic material (other than algae) could have common appearance and spectral

30

signatures as true algae. This was evident across all images, where surrounding trees were

identified as algae (abw). Everitt et al (1999) noted that water alters spectral information and

interpretation within remotely sensed imagery.

Since the images are 3-band RGB, pixel blending occurred from common spectral

information across different features, as noted between algae (abw) and tress. Picture

resolution contributed to varying classification results and accuracies. Resolution and

distortions from both the equipment and natural events (i.e., sun glints, wind, cloud cover, etc.)

impacted the programs ability classification against ground-truth features. Prevailing weather

conditions not only impact image quality but can impede the spectral measurements that can

be collected by the sensor (Hestir et al 2008).

The general understanding is that when more algal biomass is present, WQ conditions

will tend to decrease owing to respiration dominating the oxygen budget within the system.

However, images from the July to Oct indicated some of the lowest algal coverages,

comparatively (Table 1). The corresponding WQ data (Figure 11) showed increased hypoxic

and anoxic events during 2018, not following the assumption of more mass equating to more

oxygen use. This data suggests that more is occurring, likely at shorter periods than the

currently used (monthly) imagery can measure, altering the oxygen budget at the NM location,

deteriorating WQ conditions and ecologic impacts due to eutrophication within the system.

More research is needed to understand if algal coverage is accounting for the larger portion of

DO fluctuations or if other processes are contributing to the varying oxygen levels throughout

the wet/dry seasons on the Central California Coast.

31

V. CONCLUSIONS

ENVI performed well compared to hand-analysis techniques in Drone Deploy, using polygons

to identify regions or interest. Both methods can be used for estimating the coverage or extent

of potential impacts to estuarine health. ENVI provides the capability to quickly classify

remotely sensed images with 90% accuracy. Some distortion and confusion within pixel

identification, contrast, and confusion occurred reducing accuracy. The key to good

classification performance is well-developed ROIs. Images with better resolution and

definitive desired features will tend to distinguish and classify with less error.

ENVI’s ability to accurately classify imagery provides a useful tool for environmental

monitoring field, particularly looking at proxies of prevailing WQ conditions such as algae.

The capabilities to identify algal coverage from spectral information with high accuracy (Table

6) is truly a significant accomplishment considering that most algal coverage analysis is

estimates. The one true accurate method for identifying algal coverage is with in-situ water

sample collections and calculation (Katz et al 2018).

This spectral analysis tool enables large environments to be monitored and assessed in

relatively short time periods. Imagery analysis, through ENVI can be performed on the scale

of hours vice days. Use of ENVI enhances the frequency of monitoring and assessment

opportunities as well. Remotely sensed data can be acquired from field site visits or satellites

over remote, hard to access locations. ENVI could be considered more accurate due to the

spectral nature of the analysis, benefiting coastal assessments and monitoring programs. The

benefit from the use of a spectral analysis tool is to reduce the amount of time dedicated to

hand (user intensive) analysis of the images.

The value of remote sensing for assessing and managing wetlands is well established

(Everitt et al 1999). Previous works have shown that algae and phytoplankton are good

indicators for use due the visible signature of Chl-a, around 440 nm and a plankton-specific

pigment, phycocyanin; which absorbs at 620 nm (Matthews and Bernard 2013). ENVI

performed very well identifying algae with minimal misclassification due to its ability to spot

the peaks the above-mentioned ranges.

32

Future work may include more frequent flights over impacted estuaries and coastal

waters. Algae impacts and response times to environmental stressors is relatively short, on the

scale of hours to days (Sanford et al 1990). Daily to weekly flights coupled with the faster

image processing time could highlight the short response times that lead to algal blooms and

water quality degradation.

ENVI has a high potential for performing multi and hyperspectral classification studies.

Multi and hyper-spectral imaging provides the potential for collecting enhanced spatial and

spectral resolution, better understand optically complex aquatic ecosystems (Ryan et 2014, and

Xi et al 2015). Figure 13 shows the potential channel manipulations capability within ENVI.

Images were collected over NM from a Micasense Altum-M camera on a DJI Inspire UAV. Band 1 =

Red channel, Band 2 = Green, Band 3 = Blue, Band 4 = RedEdge (707 nm – 727 nm), and Band 5 NIR

(800 nm -880 nm) (Mielke, 2019). Left image is near true color looking nadir on algae. Center image

is the same area, changing channels to RedEdge, red and green. Right image is looking over trees,

water and algae (in the lower right image corner) with NIR, RedEdge, and blue channels.

Figure 13. Band Manipulation Features with ENVI

This increased spectral analysis capability may potentially increase accuracy and

feature definition of desired regions. ENVI is equipped to handle this task. Mielke (2019)

performed a study into the ability to train the ENVI’s neural networking to classify varying

33

bottom types and beach features, increasing ENVI’s standard classification percentage.

ENVI’s accuracy, capability and use in conjunction with WQ data and models may be useful

for identifying and tracking degrading conditions or coastal areas and inland aquatic

environments.

34

LIST OF REFERENCES

Ali, K.A., 2011. Prediction of Water Quality Parameters from VIS-NIR Radiometry: Using Lake

Erie as a Natural Laboratory for Analysis of Case 2 Waters. Kent State Dissertation, 1-

243

Bricker, S.B., B. Longstaff, W. Dennison, A. Jones, K. Boicourt, C. Wicks, and J. Woerner.,

2007. Effects of nutrient enrichment in the nation’s estuaries: A decade of change

[Internet]. NOAA Coastal Ocean Program Decision Analysis, Series 26. Silver Spring

(MD): NOAA, National Centers for Coastal Ocean Science. Available from

http://ccma.nos.noaa.gov/publications/eutroupdate/

Clark, R., and O’Connor, K., 2019. A systematic survey of bar-built estuaries along the

California coast. Estuarine, Coastal and Shelf Science. 226:106285.

Chuwen, B.M., Hoeksema, S.D., and Potter, I.C., 2009. The divergent environmental

characteristics of permanently-open, seasonally-open, and normally-closed estuaries of

south-western Australia. Estuarine, Coastal and Shelf Science. 85:12-21.

Cooper, J.A.G., 2001. Geomorphological variability among microtidal estuaries from the wave-

dominated South African coast. Geomorphology. 40:99-122

DroneDeploy, Program for Drone flight planning and image analysis (trial version). Accessed 09

May2020, https://www.dronedeploy.com/

Elkhorn Slough National Estuarine Research Reserve (ESNERR) website, accessed 07

November 2019, https://www.elkhornslough.org/story/

Egerton, T.A., et al, 2014. Emergence of Algal Blooms: The Effects of Short-Term Variability in

Water Quality on Phytoplankton Abundance, Diversity, and Community Composition in

a Tidal Estuary. Microorganisms 2:33-57.

Everitt., J.H., Yang, C., Escobar, D.E., Webster, C.F., Lonard, R.I., and Davis, M.R., 1999.

Using Remote Sensing and Spatial Information Technologies to Detect and Map Two

Aquatic Macrophytes. Journal of Aquatic Plant Management., 37:71-80.

Geyer, N.L., Huettel, M., and Wetz, M.S., 2018. Phytoplankton Spatial Variability in the River-

Dominated Estuary, Apalachicola Bay, Florida. Estuaries and Coasts. 41:2024-2038.

Guimaraes, T.T., Veronez, M.R., Koste, E.C., Gonzaga, L. Jr., Bordin, F., Inocencio, L.C.,

Larocca, A.P., de Oliveira, M.Z., Vitti, D.C., and Mauad, F.F., 2017. An Alternative

Method of Spatial Autocorrelation for Chlorophyll Detection in Water Bodies Using

Remote Sensing. Sustainability 9(416):1-14

Hansen, C.H., Burian, S.J., Dennison, P.E., and Williams, G.P., 2017. Spatiotemporal

Variability of Lake Water Quality in the Context of Remote Sensing Models. Remote

Sens. 9:409, https://doi.org/10.3390/rs9050409

35

Hansen, D.V., and Rattray, M. Jr., 1966. New Dimensions in Estuary Classification. Limno.

Oceanogr. 11(3):319-326

Handler, N.B., Paytan, A., Higgins, C.H., Luthy, R.G., and Boehm, A.B., 2006. Human

Development is lInked to Multiple Water Body Impairments Along the California Coast.

Estuaries and Coasts., 29:860-870.

Harris Geospatial Solutions, 2019: Calculating Confusion Matrices. Accessed 13 July 2020,

https://www.harrisgeospatial.com/docs/CalculatingConfusionMatrices.html

Harris1 Geospatial Solutions, 2019: Calculating Confusion Matrices. Accessed 10 Feb 2020,

https://www.harrisgeospatial.com/docs/SpectralAngleMapper.html

Hestir, E.L., Khanna, S., Andrew, M.E., Santos, M.J., Viers, J.H., Greenberg, J.A., Rajapakse,

S.S., and Ustin, S.L., 2008. Identification of invasive vegetation using hyperspectral

remote sensing in the California Delta ecosystem. Remote Sensing Environment.,

112:4034-4047.

Hsieh, J.L., Fries, J.S., and Noble, R.T., 2007. Vibrio and Phytoplankton Dynamics during the

Summer of 2004 in a Eutrophying Estuary. Ecological Applications. 17(5)S102-S109.

Hoeksema, S.D., Chuwen, B.M., Tweedley, J.R., and Potter, I.C., 2018. Factors influencing

marked variation in the frequency and timing of bar breaching and salinity and oxygen

regimes among normally-closed estuaries. Estuarine, Coastal and Shelf Science.,

208:205-218.

Hubertz, E.D., and Cahoon, L.B., 1999. Short-term Variability of Water Quality Parameters in

Two Shallow Estuaries of North Carolina. Estuaries., 22:814-823.

Jeppesen, R., Rodriguez, M., Rinde, J., Haskins, J., Hughes, B., Mehner, L., and Wasson, K.,

2016. Effects of Hypoxia on Fish Survival and Oyster Growth in a Highly Eutrophic

Estuary. Estuaries and Coasts., 41:89-98.

Karr, J.R. and Chu, E.W., 1999. Restoring life in running waters: Better biological monitoring.

Washington D.C.: Island Press.

Katz, C., Rivera-Duarte, I., Sorenson, K., and Chadwick, B., 2018. Santa Margarita Estuary

Water Quality Monitoring Data. Technical Report (3125 - Februray 2018)., 1-135.

Kennish, M.J., 2002. Environmental threats and environmental future estuaries. Environmental

Conservation., 29(1):78-107.

Koparan, C., Koc, A.B., Privette, C.V., and Sawyer, C.B., 2018. In Situ Water Quality

Measurements Using an Unmanned Aerial Vehicle (UAV) System. Water. 10:264,

https://doi.org/10.3390/w10030264

Kotsedi, D., Adams, J.B., and Snow, G.C., 2012. The response of microalgal biomass and

community composition to environmental factors in the Sundays Estuary. Water SA.

38:177-189.

36

Le, C., Li, Y., Zha, Y., Sun, D., Huang, C., and Zhang, H., 2011Remote estimation of

chlorophyll a in optically complex waters based on optical classification. Remote

Sensing of Environment. 115:725-737.

Le, C., Hu, C., English, D., Cannizzaro, J., Chen, Z., Lovach, C., Anastasiou, C.J., Zhao, J., and

Carder, K.L., 2013. Inherent and apparent optical properties of the complex estuarine

waters of Tampa Bay: What controls light? Estuarine, Coastal and Shelf Science 117:54-

69.

Leonardi, N., Carnacina, I., Donatelli, C., Ganju, N.K., Plater, A.J., Schuerch, M., Temmerman,

S., 2018. Dynamic interactions between coastal storms and salt marshes: A review.

Geomorphology 301:92-107

MATLAB, 2020. R2018b. Natick, Massachusetts: The MathWorks Inc., accessed 18 May 2020.

Matthews, M.W. and Bernard, S., 2013. Characterizing the Absorption Properties for Remote

Sensing of Three Small Optically-Diverse South African Reservoirs. Remote Sensing

5:4370-4404, https://doi.org/10.3390/rs5094370

Maznah, W. and Omar, W., 2010. Perspectives on the Use of Algae as Biological Indicators for

Monitoring and Protecting Aquatic Environments, with Special Reference to Malaysian

Freshwater Ecosystems. Tropical Life Sciences Research. 21(2):51-67.

McCormick, P.V., and Cairns, J. Jr., 1994. Algae as indicators of environmental change. Journal

of Applied Phycology. 6:509-526.

Mercado, L., Haskins, J., Preisler, R., 2014. A water quality report card for the Elkhorn Slough

estuary in 2013. Technical Report Series 2014:1. Available at

http://www.elkhornslough.org/research/bibliography_tr.htm

Mielke, A., and Orescanin, M.S., 2019: Using Deep Convolutional Neural Networks to Classify

Littoral Areas with 3-Band and 5-Band Imagery. M.S. Thesis, Depth of Oceanography,

Naval Postgraduate School, 64 pp

Millie, D.F., Carrick, H.J., Doering, P.H., and Steidinger, K.A., 2004. Intra-annual variability of

water quality and phytoplankton in the North Fork of the St. Lucie River Estuary, Florida

(USA): a quantitative assessment. Estuarine, Coastal and Shelf Science. 61:137-149.

NOAA National Estuarine Research Reserve System (NERRS). System-wide Monitoring

Program. Data accessed from the NOAA NERRS Centralized Data Management Office

website: http://cdmo.baruch.sc.edu/; accessed 12 October 2012.

Orescanin, M.M., Hamilton, R.P., and Hoffnagle, B., 2019. Tidal choking in an

anthropologically modified salt marsh estuary: Improving circulation through constriction

removal. Estuarine, Coastal and Shelf Science., 218:148-162,

https://doi.org/10.1016/j.ecss.2018.12.011

37

Paerl, H.W., 1998. Ecosystem responses to internal and watershed organic matter loading:

consequences for hypoxia in the eutrophying Neuse River Estuary, North Carolina, USA.

Marine Ecology Progress Series., 166:17-25

Paerl, H.W., Valdes-Weaver, L.M., Joyner, A.R., and Winkelmann, V., 2007. Phytoplankton

Indicators of Ecological Change in Eutrophying Pamlico Sound System, North Carolina.

Ecological Applications. 17(5):88-101.

Palmer, S.C., Pelevin, V.V., Goncharenko, I., Kovacs, A.W., Zlinszky, A., Presing, M., Horvath,

H., Nicolas-Perea, V., Balzter, H., and Toth, V.R., 2013. Ultraviolet Fluorescence

LiDAR (UFL) as a Measurement Tool for Water Quality Parameters in Turbid Lake

Conditions. Remote Sensing 5:4405-4422, https://doi.org/10.3390/rs5094405

Palmer, T.A., Montagna, P.A., Pollack, J.B., Kalke, R.D., and Deyoe, H.R., 2011. The role of

freshwater inflow in lagoons, rivers, and bays. Hydrobiologia. 667:49-67

Plew, D.R., Zeldis, J.R., Shankar, U., and Elliot, S., 2015: CLUES Estuary, a tool for predicting

estuary water quality. Australasian Coasts & Ports Conference 2015, Auckland, New

Zealand.

Ryan, J.P., Davis, C.O., Tufillaro, N.B., Kudela, R.M., and Gao, B., 2014. Application of the

Hyperspectral Imager for the Coastal Ocean to Phytoplankton Ecology Studies in

Monterey Bay, CA, USA. Remote Sensing., 6:1007-1025,

https://doi.org/10.3390/rs6021007

Reif, M., 2011. Remote Sensing for Inland Water Quality Monitoring: A U.S. Army Corps of

Engineers Perspective. Final Report (ERDC/EL TR-11-13), 1-44.

Riegl, B.M, and Purkis, S.J., 2005. Detection of shallow subtidal corals from IKONOS satellite

and QTC View (50, 200 kHz) single-beam sonar data (Arabian Gulf, Dubai, UAE).

Remote Sensing of Environment. 95:96-114, https://doi.org/10.1016/j.rse.2004.11.016

Rowland, G.P., Purkis, S.J., and Riegl, B.M., 2008. The 2005 Coral-bleaching Event Roatan

(Honduras): Use of Pseudo-invariant Features (PIFs) in Satellite Assessments. J. Spatial

Science. 53(1):99-112, https://doi.org/10.1080/14498596.2008.9635139

Sanford, L.P., Sellner, K.G., and Breitburg,D.L., 1990: Covariability of dissolved oxygen with

physical processes in the summertime Chesapeake Bay. J. Mar Sci. 48:567-590.

Schmit, K.L., 2010: Environmental trigger of hypoxia in the Elkhorn Slough Reserve. Civil and

Environmental Engineering, Stanford University, 9 pp.

Sharp, Jonathan H., 2010. Estuarine oxygen dynamics: What can we learn about hypoxia from

long-time records in the Delaware Estuary? Limno. Oceanogr. 51:535-548.

Stanley, D.W., 1993. Long-Term Trends in Pamlico River Estuary Nutrients, Chlorophyll,

Dissolved Oxygen, and Watershed Nutrient Production. Water Resources Research:

29:2651-2662.

38

Twomey, L., and Thompson, P., 2001. Nutrient Limitation of Phytoplankton in a Seasonally

Open Bar Built Estuary: Wilson Inlet, Western Australia. Journal of Phycology. 37:16-

29.

Weather Underground website, accessed 13 July 2020,

https://www.wunderground.com/dashboard/pws/KCAWATSO38?cm_ven=localwx_pws

dash

Wise, M.M., 2017. Nutrient Dynamics in Tidally Restricted Regions of the Elkhorn Slough

National Estuarine Research Reserve. M.S. Thesis, California State University –

Monterey Bay, 119 pp.

Xi, H., Hieronymi, M., Rottgers, R., Krasemann, H., and Qiu, Z., 2015. Hyperspectral

Differentiation of Phytoplankton Taxonomic Groups: A Comparison between Using

Remote Sensing Reflectance and Absorption Spectra. Remote Sensing., 7:14781-14805,

https://doi.org/10.3390/rs71114781